For business applications such as SAP, the journey to the cloud starts with virtual infrstructure and today approximately 30-40% of all SAP customers running SAP in production virtualized and the number is growing.
A virtual infrastructure consists of clusters of physical compute servers each running a special operating system known as a "hypervisor". Common examples of hypervisors include vSphere, Hyper-V, Xen and KVM. The hypervisor creates virtual copies of the servers hardware components such as CPU, RAM, Disk, etc. and groups these virtual HW components into a "virtual computer" aka. "virtual machine" (VM). Each server can support running dozens of VMs each running a virtualized application atop its own discrete operating system. The hypervisor will store the VM data in files either directly on a file share or it will format a file system atop a block device from a storage platform. The choice to go with a file or block storage platform is one of IT's preference more than one of technical advantages. With any VM consisting of a few large files, cloning an entire VM and it's application becomes straight forward and desirable. VI workload data duplication rates will increase with each operational replica VM used for backup, testing or reporting.
The data streams from these virtualized applications running in VMs is being multiplexed by the hypervisor before it arrives at the storage platform in a randomized state. A mixed data stream can be best be serviced by flash technology with its millisecond speeds being independent of a block's location.
Due to the workload's consolidation of many virtualized applications, there is a desire to maximize $/GB and $/IOPS as well as match the storage platform's ability to scale quickly to the speed at which virtual environments can grow. Keeping virtual infrastructure costs low will rely on efficiencies realized through integration with the virtualization software layer and possibly higher cloud automation layers. The virtual infrastructure workload requirements can be met with several of the storage architectures.
The Spiderchart below shows the distribution and weighting of the primary workload requirements for this use case.